CN100359532C - Multiple target image hierarchical clustering method - Google Patents
Multiple target image hierarchical clustering method Download PDFInfo
- Publication number
- CN100359532C CN100359532C CNB031289185A CN03128918A CN100359532C CN 100359532 C CN100359532 C CN 100359532C CN B031289185 A CNB031289185 A CN B031289185A CN 03128918 A CN03128918 A CN 03128918A CN 100359532 C CN100359532 C CN 100359532C
- Authority
- CN
- China
- Prior art keywords
- target
- layer
- point
- center
- identification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Abstract
The present invention relates to a layering and clustering method for multiple target images. Target points are arranged from big target point to small target point in the height direction first, and targets are separated in layers with equal values according to the maximum value and the minimum value of target heights to obtain all target points of each layer and to find out the corresponding relation of the target points on an upper and a lower layers; target clustering and identification are successively and singly carried out for each layer from top to bottom; finally, all target center points of each layer are combined to carry out multi-layer target information combination. The present invention can separate and identify multiple moving targets under the crowed condition with high accuracy, and the present invention is also suitable for conditions without crowed targets and the identification for multiple moving targets of video images.
Description
Technical field:
The present invention relates to a kind of hierarchical cluster method of multiple goal image, a plurality of detected target under the crowed condition is carried out hierarchical cluster and identification, belong to computerized information Flame Image Process and mode identification technology, can be applied to intelligent transportation field and other association areas.
Background technology:
It is a lot of at present the target in the image to be carried out the method for identification, according to different application backgrounds diverse ways is arranged.These methods all are the discrimination methods that carries out according to behind the single Threshold Segmentation image, and therefore very strong specific aim and limitation are arranged.
The complicacy of identification of targets and discrimination method depends on the congested conditions of target in the image, in image under the more sparse situation of target, utilize single threshold value to the target in the image (in laser image, object height is the true altitude of object, in general pattern, then be gray-scale value) cut apart, adopt ripe relatively discrimination method that the impact point of cutting apart is carried out cluster identification, can obtain effect preferably.But under the situation that target is crowded, crowded a plurality of targets can be joined together to form the connected region of target in the image, be difficult to be divided into impact point independently individual and to carry out the identification of success with the method for carrying out identification after the single layering, present ripe relatively cutting techniques all is to carry out at sparse target, the method that changes single threshold value also has a lot, as Region Segmentation method etc., but do not see the relevant algorithm of target under the crowed condition being cut apart identification at present, crowded target is carried out accurate identification become current a great problem, this is the problem that institute must solution when identification technique was applied in practical matter.
Summary of the invention:
The objective of the invention is at the deficiencies in the prior art, a kind of hierarchical cluster method of multiple goal image is provided, multiple goal is cut apart and identification problem under the solution crowed condition.
For realizing such purpose, considered the elevation information of target in the technical scheme of the present invention, the multilayer Threshold Segmentation is carried out in the target area, equivalence is divided each layer equally on the object height direction, every layer is carried out cluster and identification separately, and the target information of carrying out multilayer then merges.Determine concrete threshold value according to the application scenario of reality, obtain target all impact points in this threshold range, impact point is carried out from big to small arrangement by short transverse, obtain maximal value and minimum value, carry out five equilibrium according to maximal value and minimum value, by upper strata to lower floor equal portions threshold value space successively, threshold value is stored in the array.Cut apart the central point that obtains target according to threshold value to every layer, successively each layer carried out the identification of target from top to bottom.From the superiors, judge whether the target in this layer exists corresponding target's center's point in the target area scope of following one deck.If do not exist, then think new target; If, the distance of the central point of judgement and this layer target, then merging at these 2 as if satisfying condition is a target, otherwise thinks two targets.And the like, layer is finished the target identification to each layer successively from top to bottom.Final center and the target number that obtains all targets.
Method of the present invention is specific as follows:
1. according to the threshold value of initial setting all impact points are carried out from big to small arrangement, obtain the maximal value H of object height
MaxWith minimum value H
Min
2. according to the maximal value of object height and minimum value target is carried out equivalence level and cut apart, set according to concrete application scenario between cut section, obtain all impact point of each layer.
3. whether target's center's point of judging the n layer in the target area of n+1 layer, travels through all target's center's points.If the bee-line (the breadth extreme estimated value of target) between satisfying at 2 is then thought same target.Otherwise think different targets, this target's center's point is labeled as new impact point, be stored in the corresponding array.
4. after above-mentioned steps is finished bilevel impact point corresponding relation, after corresponding target cut apart and merge,, then merge, otherwise cut apart if satisfy the condition of bee-line.The impact point that central point is not mated is thought the central point of fresh target.
5. from the superiors successively, every calculating and judgement of carrying out above-mentioned algorithm between two-layer, until each layer target's center point unified one deck to the end, at last with the position of target's center as last testing result.
Because target detected impact point (target area) under crowded condition forms connected region easily, Threshold Segmentation and the identification algorithm of using individual layer are difficult to the target in the connected region is cut apart and identification accurately, easily with the target omission in the connected region with inaccurately detect the center of target.The present invention carries out the multilayer Threshold Segmentation according to object height information to the target area, detect the impact point isolated area of target different Threshold Segmentation layers on short transverse, utilize iteration self-organization data analysis algorithm that the impact point of differing heights layer is carried out identification, the high precision that realizes multiple mobile object under the crowed condition is cut apart and identification, can not omission and flase drop and detect the central point and the number of a plurality of crowded targets exactly not.
The present invention is equally applicable to the not crowded condition of target, video image is comprised the multiple mobile object identification of gray level image and coloured image and infrared image has reference function too.
Description of drawings:
Fig. 1 is multilayer threshold value cluster of the present invention and identification synoptic diagram.
Parallelogram among Fig. 1 is represented individual-layer data, and elliptic region is wherein represented impact point, and the mean value of oval internal object point is represented target's center, layering down more, and elliptic region is big more, because threshold value is low more, impact point is many more.
Fig. 2 is that individual layer is cut apart the result with identification.
Actual in the dashed rectangle among Fig. 2 is five targets, but only identifies three targets.
Fig. 3 is cut apart result with identification for multilayer.
The mark that is marked among Fig. 3 is identical with Fig. 2.Clearly show five targets and its movement locus of identification among the figure.
Fig. 4 is that the multilayer under the crowed condition is cut apart the result with identification.
Embodiment:
Below in conjunction with concrete test figure and accompanying drawing technical scheme of the present invention is further described.
Input data of the presently claimed invention are laser scanning datas, comprise target depth L, angle of deflection, and three parameters of angle of pitch β utilize the coordinate transformation formula to convert the original laser scan-data to the coordinate X of target under earth axes, Y, Z.After each analyzing spot of target finished the conversion of coordinate, can obtain position under earth axes of each analyzing spot on the target (X, Y) and height (Z).
The present invention adopts the concrete implementation step of many Threshold Segmentation and identification algorithm as follows:
1) with the laser scanning system real time scan to raw data change into the coordinate figure X of impact point under earth axes, Y, Z obtains target at these all impact points more than threshold value according to the threshold value of initial setting, and carry out from big to small arrangement, obtain the maximal value H of object height
MaxWith minimum value H
Min
2) according to the dividing layer number N that sets, the just number of threshold value.Object height is carried out equivalence from peak to peak to be cut apart.Obtain the impact point (target area) of target, as shown in Figure 1 at each Threshold Segmentation layer.Parallelogram among the figure is represented the Threshold Segmentation layer, the impact point zone that little oval representative is wherein obtained.
3) impact point that each layer is partitioned into carries out iteration self-organization data automatic cluster, determines that the central point of each target is represented target, preserves all impact points that belongs to same target simultaneously.Be recorded in the Dynamic Array cutting apart with identification result of each layer.In Fig. 1, use the central point of oval central point signal target.
4) merge according to the target's center's point and the impact point of algorithm steps noted earlier to each layer, whether target's center's point between judgement is whenever two-layer up and down and impact point zone have covering or comprise.And judge whether that new impact point occurs.Target's center's point of each layer is carried out again calculating and judgement.
5) with the unified one deck to the end of each layer target's center's point, the target's center point that guarantees each layer is not omitted and is not repeated.The position of target's center is exported as last testing result.Be illustrated in figure 3 as the result of multilayer Threshold Segmentation.
Accompanying drawing 2 and accompanying drawing 3 are that individual layer is cut apart and is segmented in target under the same moving scene with multilayer and cuts apart result with identification.Fig. 4 is that the multilayer under the crowed condition is cut apart the result with identification, target in the border, field is very crowded, actual range between the target can find out that from result shown in the drawings algorithm that multilayer is cut apart can detect target location and the number under the crowed condition fully exactly greatly about about 20cm.
Claims (1)
1, a kind of hierarchical cluster method of multiple goal image is characterized in that comprising the steps:
1) according to the threshold value of initial setting all impact points is carried out from big to small arrangement, obtain the maximal value H of object height
MaxWith minimum value H
Min
2) according to the maximal value of object height and minimum value target is carried out equivalence level and cut apart, set according to concrete application scenario between cut section, obtain all impact point of each layer;
3) judge that target's center's point of n layer is whether in the target area of n+1 layer, travel through all target's center's points, if the bee-line between satisfying at 2 is the breadth extreme estimated value of target, then think same target, otherwise think different targets, this target's center's point is labeled as new impact point, is stored in the corresponding array;
4) after finishing bilevel impact point corresponding relation, corresponding target is cut apart and merged, if satisfy the condition of bee-line, then merge, otherwise cut apart, the impact point that central point is not mated is thought the central point of fresh target;
5) from the superiors successively, until with the unified one deck to the end of each layer target's center's point, with the position of target's center as last testing result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB031289185A CN100359532C (en) | 2003-05-29 | 2003-05-29 | Multiple target image hierarchical clustering method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB031289185A CN100359532C (en) | 2003-05-29 | 2003-05-29 | Multiple target image hierarchical clustering method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1452130A CN1452130A (en) | 2003-10-29 |
| CN100359532C true CN100359532C (en) | 2008-01-02 |
Family
ID=29222925
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB031289185A Expired - Fee Related CN100359532C (en) | 2003-05-29 | 2003-05-29 | Multiple target image hierarchical clustering method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100359532C (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP4412342B2 (en) * | 2007-03-30 | 2010-02-10 | ソニー株式会社 | CONTENT MANAGEMENT DEVICE, IMAGE DISPLAY DEVICE, IMAGING DEVICE, PROCESSING METHOD IN THEM, AND PROGRAM FOR CAUSING COMPUTER TO EXECUTE THE METHOD |
| CN101339652B (en) * | 2007-12-28 | 2011-06-01 | 中国人民解放军海军航空工程学院 | Solid engines CT image division method |
| CN101216886B (en) * | 2008-01-11 | 2010-06-09 | 北京航空航天大学 | A shot clustering method based on spectral segmentation theory |
| CN101702236B (en) * | 2009-10-30 | 2011-09-21 | 无锡景象数字技术有限公司 | Multi-target foreground segmentation method |
| CN102799667B (en) * | 2012-07-13 | 2015-04-29 | 北京工商大学 | Hierarchical clustering method based on asymmetric distance |
| CN104244035B (en) * | 2014-08-27 | 2018-10-02 | 南京邮电大学 | Network video stream sorting technique based on multi-level clustering |
| CN110807807B (en) * | 2018-08-01 | 2022-08-05 | 深圳市优必选科技有限公司 | Monocular vision target positioning pattern, method, device and equipment |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5379044A (en) * | 1993-12-23 | 1995-01-03 | Hughes Aircraft Company | Efficient multi-target tracking method |
| JPH0822534A (en) * | 1994-07-11 | 1996-01-23 | Fujitsu Ltd | Multitarget tracking system |
| CN1361503A (en) * | 2000-12-29 | 2002-07-31 | 南开大学 | Color multi-objective fusion identifying technology and system based on neural net |
| CN1367468A (en) * | 2002-03-25 | 2002-09-04 | 北京工业大学 | Cornea focus image cutting method based on K-mean cluster and information amalgamation |
| CN1371504A (en) * | 1999-01-13 | 2002-09-25 | 电脑相关想象公司 | Signature recognition system and method |
| CN1389808A (en) * | 2002-07-18 | 2003-01-08 | 上海交通大学 | Multiple-moving target tracking method |
| CN1389710A (en) * | 2002-07-18 | 2003-01-08 | 上海交通大学 | Multiple-sensor and multiple-object information fusing method |
-
2003
- 2003-05-29 CN CNB031289185A patent/CN100359532C/en not_active Expired - Fee Related
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5379044A (en) * | 1993-12-23 | 1995-01-03 | Hughes Aircraft Company | Efficient multi-target tracking method |
| JPH0822534A (en) * | 1994-07-11 | 1996-01-23 | Fujitsu Ltd | Multitarget tracking system |
| CN1371504A (en) * | 1999-01-13 | 2002-09-25 | 电脑相关想象公司 | Signature recognition system and method |
| CN1361503A (en) * | 2000-12-29 | 2002-07-31 | 南开大学 | Color multi-objective fusion identifying technology and system based on neural net |
| CN1367468A (en) * | 2002-03-25 | 2002-09-04 | 北京工业大学 | Cornea focus image cutting method based on K-mean cluster and information amalgamation |
| CN1389808A (en) * | 2002-07-18 | 2003-01-08 | 上海交通大学 | Multiple-moving target tracking method |
| CN1389710A (en) * | 2002-07-18 | 2003-01-08 | 上海交通大学 | Multiple-sensor and multiple-object information fusing method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1452130A (en) | 2003-10-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110175576B (en) | Driving vehicle visual detection method combining laser point cloud data | |
| CN106650640B (en) | Negative obstacle detection method based on laser radar point cloud local structure characteristics | |
| EP3471026A1 (en) | Method for acquiring bounding box corresponding to an object in an image by using convolutional neural network including tracking network and computing device using the same | |
| US8670612B2 (en) | Environment recognition device and environment recognition method | |
| KR101268523B1 (en) | Fast scene understanding method in urban environment using laser scanner | |
| US9099005B2 (en) | Environment recognition device and environment recognition method | |
| CN109633674A (en) | Three-dimensional Track automatic planning is maked an inspection tour in transmission of electricity based on laser point cloud data | |
| US9075026B2 (en) | Defect inspection device and defect inspection method | |
| CN109633676A (en) | A kind of method and system based on the laser radar obstruction detection direction of motion | |
| CN106204547B (en) | The method that rod-shaped atural object spatial position is automatically extracted from Vehicle-borne Laser Scanning point cloud | |
| CN108062517A (en) | Unstructured road boundary line extraction method based on vehicle-mounted laser point cloud | |
| US20120288150A1 (en) | Environment recognition device and environment recognition method | |
| CN100359532C (en) | Multiple target image hierarchical clustering method | |
| CN107038442A (en) | A kind of car plate detection and global recognition method based on deep learning | |
| CN109870458B (en) | Pavement crack detection and classification method based on three-dimensional laser sensor and bounding box | |
| CN113759391A (en) | Passable area detection method based on laser radar | |
| CN102881017B (en) | Cell separation method | |
| CN110554409B (en) | Concave obstacle detection method and system | |
| CN111640323A (en) | Road condition information acquisition method | |
| CN108074232A (en) | A kind of airborne LIDAR based on volume elements segmentation builds object detecting method | |
| CN111860137A (en) | Track turnout identification method based on vision | |
| AU2022258555B2 (en) | Generating mappings of physical spaces from point cloud data | |
| KR101000332B1 (en) | Drivable road detecting method and outdoor driving robot using the same | |
| CN114332823A (en) | Drivable area detection method based on semantic segmentation | |
| CN112749741B (en) | Hand brake fastening fault identification method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20080102 Termination date: 20100529 |